Files and links (1)
Free to read and download, Open Access
Thesis
Segmentation of volumetric medical image data using deep convolutional neural networks
University of Iowa
Master of Science (MS), University of Iowa
Summer 2018
DOI: 10.17077/etd.006120
Abstract
Tomographic imaging plays an important role in medical decision making. For many applications, image segmentation is needed to perform quantatative analysis of medical image data. The goal of this work was to assess the utility of deep convolutional neural networks for volumetric image data segmentation. For this purpose several deep convolutional neural network frameworks and network architectures were investigated and compared in the context of cerebellum segmentation in volumetric head and neck PET-CT scans. Our results showed that the Isensee network architecture produced the most accurate segmentations, and that the utilization of an exponential learning rate decay combined with data augmented for training were beneficial to the performance of the network. We determined that the CT channel of the PET-CT images was not beneficial enough to warrant its inclusion in the training process. In addition, the feasibility of lung segmentation in volumetric CT scans was demonstrated and assessed. In summary, the selected deep convolutional neural network approach represents a suitable approach for volumetric medical image data segmentation.
Details
- Title: Subtitle
- Segmentation of volumetric medical image data using deep convolutional neural networks
- Creators
- Timothy John Linhardt
- Contributors
- Reinhard R Beichel (Advisor)Andreas Wahle (Committee Member)Brian J Smith (Committee Member)
- Resource Type
- Thesis
- Degree Awarded
- Master of Science (MS), University of Iowa
- Degree in
- Electrical and Computer Engineering
- Date degree season
- Summer 2018
- Publisher
- University of Iowa
- DOI
- 10.17077/etd.006120
- Number of pages
- xii, 111 pages
- Copyright
- Copyright 2018 Timothy John Linhardt
- Language
- English
- Description illustrations
- illustrations (some color)
- Description bibliographic
- Includes bibliographical references (pages 106-111).
- Academic Unit
- Electrical and Computer Engineering
- Record Identifier
- 9984097173102771
Metrics
7 File views/ downloads
32 Record Views